Dynamic Range and SFDR of ADC

Hi all, 

I am using ADC to sample a sine wave. The closer the sine wave frequency
is to the half of sample rate, the more harmonics I get.

However, I would say that starting from some frequency, SFDR is lower than
the one specified in ADC specifications.

How can I then know if my received signal is good enough? Does this mean
that with certain sampling frequency I can only observe signals up till
certain signal frequency that gives expected SFDR?
If I get higher spurs (harmonics) than expected can this be related to
signal source, cabling or some other source of error in my receiving
process?
The way in which I measure SFDR is by ploting 20log10(fft(x)/length(x))
for certain discrete time sequence x[n] and then observing difference
between fundamental frequency and the strongest harmonic.

Is Dynamic range only depending on the number of bits of ADC?


---------------------------------------
Posted through http://www.DSPRelated.com

On 2016-03-14 4:25, b2508 wrote:
> Hi all, 
> 
> I am using ADC to sample a sine wave. The closer the sine wave frequency
> is to the half of sample rate, the more harmonics I get.

I guess you mean more "the higher the harmonics are"?

> However, I would say that starting from some frequency, SFDR is lower than
> the one specified in ADC specifications.

It is quite normal that SNDR/SFDR is frequency dependent (input signal
and sampling frequency) and these plots are usually in the datasheet
(e.g. page 8 of AD9203).

The "one specified in the ADC specifications" is usually measured under
certain conditions. E.g. the AD9203 mentions ENOB (which is basically
SNDR): 9.55 @ fIN = 20 MHz

> How can I then know if my received signal is good enough? Does this mean
> that with certain sampling frequency I can only observe signals up till
> certain signal frequency that gives expected SFDR?

I am not sure if I understand the question. If your signal covers
(0,...,fs/2) then you need to take the worst value on the plot above.

If your input signal covers a limited range, you see how much distortion
you can expect there.

> If I get higher spurs (harmonics) than expected can this be related to
> signal source, cabling or some other source of error in my receiving
> process?
> The way in which I measure SFDR is by ploting 20log10(fft(x)/length(x))
> for certain discrete time sequence x[n] and then observing difference
> between fundamental frequency and the strongest harmonic.

If your results roughly match the plots from the datasheet, then other
sources of error are lower. If there is a discrepancy, there may be
other sources of error present. Cabling is more unlikely in tems of
linearity but amplifiers may introduce significant distortion.

> Is Dynamic range only depending on the number of bits of ADC?

No. The actual number of bits don't say much in the first place, hence
there is a figure ENOB (effecive number of bits). It's how many bits you
get out including the error sources (quantization noise, thermal noise,
INL/DNL, ...).This figure is directly related to SNDR:

ENOB = (SNDR-1.76)/6.02

(by the way, there is a big related discussion regarding dynamic range
of an ADC in "daily quiz").


Peter

Peter Mairhofer <63832452@gmx.net> writes:
> [...]
> It is quite normal that SNDR/SFDR is frequency dependent (input signal
> and sampling frequency) and these plots are usually in the datasheet
> (e.g. page 8 of AD9203).

Peter,

What is "SNDR?"
-- 
Randy Yates, DSP/Embedded Firmware Developer
Digital Signal Labs
http://www.digitalsignallabs.com

On 2016-03-14 21:23, Randy Yates wrote:
> Peter Mairhofer <63832452@gmx.net> writes:
>> [...]
>> It is quite normal that SNDR/SFDR is frequency dependent (input signal
>> and sampling frequency) and these plots are usually in the datasheet
>> (e.g. page 8 of AD9203).
> 
> Peter,
> 
> What is "SNDR?"

Just another word for SINAD (Signal-to-Noise-And-Distortion-Ratio).

Peter

On Monday, March 14, 2016 at 7:25:23 AM UTC-4, b2508 wrote:
> 
> I am using ADC to sample a sine wave. The closer the sine wave frequency
> is to the half of sample rate, the more harmonics I get.

harmonics?  or aliases?  at what frequencies are those "harmonics"?  appears to me, how this statement is constructed that any harmonics you get must be (much) higher in frequency than Nyquist (which is half the sample rate).

> However, I would say that starting from some frequency, SFDR is lower than
> the one specified in ADC specifications.

"SFDR" = "Signal to noise and Distortion Ratio".  (what's the "F"?)


> How can I then know if my received signal is good enough? Does this mean
> that with certain sampling frequency I can only observe signals up till
> certain signal frequency that gives expected SFDR?
> If I get higher spurs (harmonics) than expected can this be related to
> signal source, cabling or some other source of error in my receiving
> process?
> The way in which I measure SFDR is by ploting 20log10(fft(x)/length(x))
> for certain discrete time sequence x[n] and then observing difference
> between fundamental frequency and the strongest harmonic.
> 
> Is Dynamic range only depending on the number of bits of ADC?

the definition of dynamic range that i advocate is the sum (in log amplitude, such as dB) of S/N ratio plus headroom.  for a single ADC, if your signal is 1/4 full scale, you have 12 dB of headroom and some determined S/N ratio.  if you double the amplitude of that signal to 1/2 full scale, you will increase your signal by 6 dB, if the ADC ain't going non-linear on you, your quantization noise should be about the same, so your S/N ratio has increased by 6 dB.  but your headroom was decreased from 12 dB to 6 dB and the sum of the two quantities is about the same and is an attribute of the ADC.  with signals at different frequencies and different amplitudes the measured dynamic range will vary a little, but it should be roughly constant, at least if you have at least 3 or 6 dB of headroom.

in this working definition, if you're careful to compare apples to apples, the dynamic range is mostly constant within the normal operating range of the ADC and is 6.02 dB per *honest* bit of ADC.  defining the number of bits of an ADC can end up in a specsmanship contest.  any 24-bit audio ADC has, in my opinion, at least 3 or 4 "marketing bits" because the dynamic range is not much better than 120 dB.

so, if i were to define the number of "honest bits" of an ADC, i would take the best measured dynamic range in dB and multiply that by 1/6 bit per dB.

now i am *not* saying that ADC manufacturers should consider the "noisy" bits below the least significant honest bit to be useless.  like with sigma-delta ADCs, the word width is growing as they do the arithmetic in the decimation process and internally they could have a *very* wide word by the time it is output.  the real output signal can be considered a sum of the noiseless input plus a noisy signal and just truncating the bits below the "honest" LSB simply adds another source of quantization noise.  but we should recognize that those low bits are noisy.  they sorta have dither added to them for free.  so i would *not* lose the bits below the 20th bit.  if it's a 32-bit DSP, i would want all 32 bits, but i will be happy for 24 bits, even if the bottom 4 bits are noisy.

r b-j

On Fri, 18 Mar 2016 23:39:28 -0700, robert bristow-johnson wrote:

> On Monday, March 14, 2016 at 7:25:23 AM UTC-4, b2508 wrote:
>> 
>> I am using ADC to sample a sine wave. The closer the sine wave
>> frequency is to the half of sample rate, the more harmonics I get.
> 
> harmonics?  or aliases?  at what frequencies are those "harmonics"? 
> appears to me, how this statement is constructed that any harmonics you
> get must be (much) higher in frequency than Nyquist (which is half the
> sample rate).
> 
>> However, I would say that starting from some frequency, SFDR is lower
>> than the one specified in ADC specifications.
> 
> "SFDR" = "Signal to noise and Distortion Ratio".  (what's the "F"?)

SFDR = Spurious Free Dynamic Range.  Basically the ratio between the 
power of the high-amplitude, narrowband test signal (or signals if doing 
a multi-tone test) and the highest power in any of the other FFT bins.
Typically this is determined by harmonic distortion, intermod. distortion 
or aliases of those.  Noise doesn't come into it.

It's a useful metric for test equipment, e.g. a spectrum analyser.

C.f. SINAD, which is more useful for communications applications.

Regards,
Allan

On 3/19/2016 3:01 AM, Allan Herriman wrote:
> On Fri, 18 Mar 2016 23:39:28 -0700, robert bristow-johnson wrote:
>
>> On Monday, March 14, 2016 at 7:25:23 AM UTC-4, b2508 wrote:
>>>
>>> I am using ADC to sample a sine wave. The closer the sine wave
>>> frequency is to the half of sample rate, the more harmonics I get.
>>
>> harmonics?  or aliases?  at what frequencies are those "harmonics"?
>> appears to me, how this statement is constructed that any harmonics you
>> get must be (much) higher in frequency than Nyquist (which is half the
>> sample rate).
>>
>>> However, I would say that starting from some frequency, SFDR is lower
>>> than the one specified in ADC specifications.
>>
>> "SFDR" = "Signal to noise and Distortion Ratio".  (what's the "F"?)
>
> SFDR = Spurious Free Dynamic Range.  Basically the ratio between the
> power of the high-amplitude, narrowband test signal (or signals if doing
> a multi-tone test) and the highest power in any of the other FFT bins.
> Typically this is determined by harmonic distortion, intermod. distortion
> or aliases of those.  Noise doesn't come into it.
>
> It's a useful metric for test equipment, e.g. a spectrum analyser.
>
> C.f. SINAD, which is more useful for communications applications.

I don't know why SFDR would not be important in comms work.  If you wish 
to pick up a weak signal in the presence of a strong signal that spur 
could be very important if it is close enough to your signal of interest.

-- 

Rick

On Sat, 19 Mar 2016 03:24:09 -0400, rickman wrote:

> On 3/19/2016 3:01 AM, Allan Herriman wrote:
>> On Fri, 18 Mar 2016 23:39:28 -0700, robert bristow-johnson wrote:
>>
>>> On Monday, March 14, 2016 at 7:25:23 AM UTC-4, b2508 wrote:
>>>>
>>>> I am using ADC to sample a sine wave. The closer the sine wave
>>>> frequency is to the half of sample rate, the more harmonics I get.
>>>
>>> harmonics?  or aliases?  at what frequencies are those "harmonics"?
>>> appears to me, how this statement is constructed that any harmonics
>>> you get must be (much) higher in frequency than Nyquist (which is half
>>> the sample rate).
>>>
>>>> However, I would say that starting from some frequency, SFDR is lower
>>>> than the one specified in ADC specifications.
>>>
>>> "SFDR" = "Signal to noise and Distortion Ratio".  (what's the "F"?)
>>
>> SFDR = Spurious Free Dynamic Range.  Basically the ratio between the
>> power of the high-amplitude, narrowband test signal (or signals if
>> doing a multi-tone test) and the highest power in any of the other FFT
>> bins. Typically this is determined by harmonic distortion, intermod.
>> distortion or aliases of those.  Noise doesn't come into it.
>>
>> It's a useful metric for test equipment, e.g. a spectrum analyser.
>>
>> C.f. SINAD, which is more useful for communications applications.
> 
> I don't know why SFDR would not be important in comms work.  If you wish
> to pick up a weak signal in the presence of a strong signal that spur
> could be very important if it is close enough to your signal of
> interest.

Every ADC parameter is always important in some way.

For a comms receiver, the performance will (very roughly) be determined 
by the total amount of unwanted stuff in the receiver bandwidth.

SFDR is a measure of the highest spur, and doesn't take the bandwidth 
into account.  Nor does it consider the total power in all the spurs 
(just the highest one), or that the frequency plan will usually put the 
worst spurs well out of band if the bandwidth is a small fraction of Fs.

OTOH, the rx bandwidth might be a large fraction of Fs, e.g. in a 
multichannel receiver, and the SFDR might indeed be the best ADC FOM to 
use.


For the lurkers, here's an ADC datasheet:
<http://www.analog.com/media/en/technical-documentation/data-sheets/
AD9652.pdf>
Figures 6 through 32 show various measurements of SFDR for one and two 
tone tests.

Regards,
Allan

On Saturday, March 19, 2016 at 7:58:51 AM UTC-4, Allan Herriman wrote:
> 
...
> 
> 
> For the lurkers, here's an ADC datasheet:
> <http://www.analog.com/media/en/technical-documentation/data-sheets/
> AD9652.pdf>

i was interested in this introductory quote:

"The AD9652 is a dual, 16-bit analog-to-digital converter (ADC)
with sampling speeds of up to 310 MSPS. It is designed to
support demanding, high speed signal processing applications
that require exceptional dynamic range over a wide input
frequency range (up to 465 MHz)."

so how does it do wideband up to 3x Nyquist?  i could understand maybe sampling a narrowband signal up there if you can keep the aliases from interfering.

did i read it wrong?

r b-j

robert bristow-johnson <rbj@audioimagination.com> writes:

> On Saturday, March 19, 2016 at 7:58:51 AM UTC-4, Allan Herriman wrote:
>> 
> ...
>> 
>> 
>> For the lurkers, here's an ADC datasheet:
>> <http://www.analog.com/media/en/technical-documentation/data-sheets/
>> AD9652.pdf>
>
> i was interested in this introductory quote:
>
> "The AD9652 is a dual, 16-bit analog-to-digital converter (ADC)
> with sampling speeds of up to 310 MSPS. It is designed to
> support demanding, high speed signal processing applications
> that require exceptional dynamic range over a wide input
> frequency range (up to 465 MHz)."
>
> so how does it do wideband up to 3x Nyquist?  i could understand maybe
> sampling a narrowband signal up there if you can keep the aliases from
> interfering.

Yeah, I would think "bandpass sampling," which is what you mean, I
believe. This is usually a measure of the front-end amplifier and/or
sample/hold bandwidth and somewhat independent of sample rate.
-- 
Randy Yates, DSP/Embedded Firmware Developer
Digital Signal Labs
http://www.digitalsignallabs.com